More Diverse, Still Segregated?
2025-12-04
Introduction
Diversity at work and its implications for recruitment (DIREK)
- Christian Albrekt Larsen
- Jeevitha Yogachandiran Qvist
Two workpackages:
How diverse are Danish workplaces across gender, age, and immigrant origin, in the period from 1996 to 2022?
How does the diversity at management and employee levels affect the recruitment of new employees, and how are these relationships conditioned by the workplace’s organization, recruitment practices, and degree of competitive exposure?
How diverse are Danish workplaces across gender, age, and immigrant origin, in the period from 1996 to 2022?
Report and interactive barometer for benchmarking
Research paper: What drives changes in workplace segregation? (todays presentation)
When the labor force becomes more diverse, does this diversity naturally “trickle down” into more diverse workplaces through competitive labor market processes (pool effects; supply-driven integration), or does different groups continue to be sorted into their own respective workplaces and labor market segments (sorting effects; demand-driven segregation).
Segregated workplaces and motivation
Similar to, yet different from, segregated occupations and/or sectors
- Sectors, occupations, and workplaces all measure the uneven distribution of demographic groups across labor market positions
- Can be quantified using similar indices (dissimilarity, isolation, entropy)
- All three have been invoked to explain wage gaps, inequality, and social stratification.
- Abstraction defined by classification schemes vs. distribution across actual physical establishments
Segregated workplaces and motivation
Core intrest is intergroup exposure in daily life
- Who works together at the same place (address) at the same time
- For contact theory, workplace segregation is what matters
- An occupation may appear integrated in aggregate while individual workplaces remain highly segregated
- Workplace segregation is not simply a more granular version of occupational segregation:
- Workers in the same workplace often occupy different occupations (nurse, administrator, janitor)
- Workers in the same occupation often work at different workplaces
Two competing frameworks
Pool effects
- Rooted in neoclassical labor market theory(ies) \(^1\)
- Discrimination is “expensive” → competitive markets erode it
- Workers seek jobs matching their skills; employers seek the best match
- Prediction: As underrepresented groups grow in the labor force, and becomes increasingly similar to the “majority”, workplace composition follows
Sorting effects
- Rooted in segmented labor market theory \(^2\)
- Demand-side mechanisms maintain segregation independent of supply
- Employers channel groups into different positions, workplaces, or sectors
- Mechanisms: Hiring preferences, network recruitment, institutional practices, self-selection \(^3\)
Which of the two dominate in the allocation of workers?
An illustration
A hypothetical and stylized example of recruitment in an high-skilled “engineering workplace” in the context of engineering graduates having changed from being 90% male to 70% male over the last 20 years
Pool effects
The workplace will “naturally” include a greater number of female engineers as the company hires from the changing graduate student pool and the demographics of the workplace will continue to reflect the demographic trends of the population over time.
Sorting effects
The company will continue to hire male engineers, because this is what they have always done, and know, and therefore female engineers will cluster in other workplaces.
Four hypotheses structure the study (and this presentation)
Hypotheses (1)
H1a (Pool Effects): Workplace segregation will decrease as labor force diversity increases, reflecting the competitive translation of supply-side demographic shifts into workplace representation.
H1b (Sorting Effects): Workplace segregation will remain stable or increase despite growing labor force diversity, reflecting demand-side mechanisms that maintain segregated employment structures independent of labor pool composition.
Hypotheses (2)
It not enough that a previously underrepresented groups becomes bigger, the group must also, on average, have the skills to be a match for different type of workplaces.
- If no men have a nursing degree, they cannot work as a nurse.
- If immigrants, on average, have less years of education, they will be less represented in high-skilled workplaces
- Skill-mediated sorting: Segregation as a by-product of skill matching
- Segregation occurs without discriminatory intent
- Residual Sorting: Segregation beyond what skills can explain (unobservables)
┌─────────────────────────┐
│ Observed Segregation │
└───────────┬─────────────┘
│
┌───────────────┴───────────────┐
▼ ▼
┌───────────────────┐ ┌───────────────────┐
│ Skill-mediated │ │ Residual sorting │
│ sorting (pools) │ │ │
├───────────────────┤ ├───────────────────┤
│ Education │ │ Discrimination │
│ Occupation │ │ Networks │
│ │ │ Preferences │
└───────────────────┘ └───────────────────┘
Supply-side Demand-side
(qualification (employer/worker
matching) behavior)Key question: How much segregation remains after accounting for skill differences?
Note: Skill refers to both level of education and type of occupation/degree
Hypotheses (2)
H2 (Skill Sorting Hypothesis): The relative contribution of skill-mediated versus residual sorting varies by demographic group. Specifically:
- Immigrants: Segregation is primarily explained by education (credential recognition, education mismatch)
- Women: Segregation is primarily explained by occupation (occupational channeling, gendering of professions/educations)
- Seniors: Segregation reflects both skill- and residual sorting (skill depreciation and age-based sorting)
Hypothesis (3)
Segregation is not only a product of who workers are—it also depends on where they are employed.
- \(H2\) focus on worker-level mechanisms (skills, networks, preferences)
- Segregation processes is expected to vary by workplace-level characteristics
- The same demographic group may experience different sorting dynamics depending on organizational context
Workplaces differ in:
- Size
- Skill composition
- Hiring formalization
- Regulatory exposure
- Turnover rates
Hypothesis (3)
H3: (Workplace Size Hypothesis): Between-workplace segregation decreases with organizational size.
- Visibility pressure: Subject to regulatory scrutiny, ESG reporting, public accountability
- More hiring decisions: Greater opportunity for workforce composition to adjust to labor pool changes
- Less network hiring: Small workplaces rely more on referrals, which reproduce existing demographics through homophily
Countervailing forces: Large workplaces have internal hierarchies that enable within-workplace segregation across departments, even when overall representation looks balanced
Hypothesis (4)
H4 (Workplace Skill-Level Hypothesis): The level of segregation varies systematically across workplace skill-levels, with group-specific patterns:
Immigrants
Prediction: U-shaped pattern
- High-skill: Credential recognition barriers \(^1\)
- Low-skill: Network-based channeling into ethnic niches \(^2\)
Women
Prediction: Inversed U-shaped pattern
- Middle-skill: Occupational devaluation and gendered labor queues \(^3\)
Seniors
Prediction: U-shaped pattern
- High-skill: Entrenched human capital, firm-specific knowledge protects position \(^4\)
- Low-skill: Displacement into “bridge jobs” with lower barriers but lower quality \(^5\)
H4: Predicted patterns by workplace skill-level
Denmark as a most likely case?
If pool effects operate anywhere, they should operate here.
Legal framework: Comprehensive anti-discrimination laws, explicit “inclusive labor market” policy
Tight labor markets: Unemployment fell from >10% (1990s) to 2–3%—employers must draw from a broad labor pool
Demographic change: Clear variation in labor pool composition
| Group | 1996 | 2022 |
|---|---|---|
| Immigrants | 3.7% | 16.4% |
| Seniors (55+) | 12.0% | 21.1% |
| Women | ~50% | ~50% |
The flexicurity model
Flexibility
- Weak employment protection
- Easy to hire and fire
Security
- Generous unemployment insurance
- Active labor market policy
- Workers can refuse bad matches
Implication: High turnover enables sorting to occur continuously if workplaces adjust composition through selective retention and termination.
Denmark provides a strong test because pool effects have every advantage:
If segregation persists despite legal protections, tight labor markets, and massive demographic shifts → sorting effects dominate
If workplaces become proportionally representative → pool effects dominate
Design
Data
Integrated Database for Labour Market Research (IDA)
- Employer-employee linked dataset covering all Danish workers
- Annual observations with accurate workplace identification
Register-based Labour Force Statistics (RAS)
- Standardized labor force classifications (international standards)
- Complements IDA with occupation and industry codes
Coverage: 1996–2022
Sample construction
Restriction: Establishments with 50+ (full-time) employees
Why large workplaces?
- Stable exposure measures (small workplaces → noisy estimates)
- Formalized HR policies and hiring practices
- ESG reporting requirements
- Higher turnover → more hiring decisions observed
N: 12,000 workplaces and 1,500,000 workers in 2022. Approximately half of the Danish workforce
Demographic classifications
| Group | Definition | Reference group |
|---|---|---|
| Gender | Female (administrative record) | Male |
| Immigrant | First generation (foreign-born, no Danish parent) + descendants (Danish-born, no Danish parent) | Danish origins (one or both parents Danish-born) |
| Senior | Age 55 or older | Age below 55 |
Skill dimensions
Three measures capture worker and workplace skill levels:
1. Education (HFUDD classification)
- Basic (ISCED 0–2)
- Vocational (ISCED 3–4)
- Short-term tertiary (ISCED 5)
- Medium-long term tertiary (ISCED 6–8)
2. Occupation (ISCO-08, 1-digit)
- 10 categories: Managers, Professionals, Technicians, Clerical, Service/Sales, Agriculture, Craft/Trades, Operators, Elementary, Armed Forces
Workplace skill index
Composite measure of workplace skill level
| Component | Weight | Source |
|---|---|---|
| Average educational attainment | 60% | HFUDD |
| Median income | 20% | IDA |
| Occupation rank | 20% | ILO ISCO-08 skill ranking |
Note: Currently limited to 1996-2007 while robostness checks is done on the construction of the measure from 2008-2022.
Measuring segregation and intergroup exposure
Step 1: Who are your coworkers?
For each worker, I calculate the share of coworkers from a given demographic group:
\[ \text{Share}_{iwt} = \frac{\sum_{j \neq i \in w} \mathbb{1}(group_j = g)}{n_w - 1} \]
Example: A 10-person workplace with 4 women
- A woman in this workplace has 3/9 = 33% female coworkers
- A man in this workplace has 4/9 = 44% female coworkers
Step 2: Isolation and Exposure
I then aggregate these shares separately for in-group and out-group members:
Isolation (I) — Average share of in-group coworkers for in-group members
\[I = \frac{1}{N_g}\sum_{i \in g} \text{Share}_{iwt}\]
Exposure (E) — Average share of in-group coworkers for out-group members
\[E = \frac{1}{N_{\neg g}}\sum_{i \notin g} \text{Share}_{iwt}\]
Example (Women in 2022, actual numbers):
- I = 63%: The average woman has 63% female coworkers
- E = 37%: The average man has 37% female coworkers
Step 3: Segregation as a gap in exposure/experience
\[ S = I - E \]
Segregation measures the gap in exposure to a group between in-group and out-group members.
Continuing the example:
- S = 63% − 37% = 26 percentage points
Interpretation:
- Women have 26 pp more female coworkers than men do
- Or: a woman is 1.70× as likely to have a female coworker as a man is (63 ÷ 37)
What does S tell us?
- S = 0 → In-group and out-group have identical coworker composition (no segregation)
- S > 0 → In-group members encounter more of their own group than out-group members do
- S = 100 → Complete segregation (in-group only works with in-group)
Step 4: Benchmarking against chance
Some segregation occurs by chance alone due to workplace size variation and geography. I simulate random assignment (fixed workplace sizes and regions) to establish a baseline.
\[ S^{E} = \frac{S^{\text{observed}} - S^{\text{simulated}}}{100 - S^{\text{simulated}}} \]
Example:
- Observed: I = 63%, E = 37%, S = 26%
- Simulated (random): I = 52%, E = 48%, S = 4%
- Maximum possible: S = 100%
\[S^{E} = \frac{26 - 4}{100 - 4} = \frac{22}{96} = 23\%\]
About a quarter of all possible excess segregation is realized.
Note: In this sample (workplaces with 50+ employees), \(S\) and \(S^E\) are nearly identical. Large workplaces produce minimal “segregation by chance”—random assignment would make coworker composition converge toward population averages. Virtually all observed segregation reflects systematic sorting, not statistical noise.
Conditional segregation analysis
To answer whether segregation is driven by skill differences or by other mechanisms, I randomly reassign workers within skill categories, holding workplace skill composition fixed: A nurse can only be assigned to workplaces that employ nurse, etc.
The gap between observed and this conditional baseline tells us how much segregation is skill-mediated. What remains after conditioning is residual
- 0% → Skills explain nothing
- 100% → Skills explain everything
Extensions
H3 (Size): Compute \(S^E\) by workplace size quintile
H4 (Skill level): Compute \(S^E\) by workplace skill quintile
Results
H1: Pool Effects vs. Sorting Effects
Does increased labor force diversity lead to more workplace diversity?
If pool effects dominate: Diversity ↑ → Segregation ↓ (negative correlation)
If sorting effects dominate: Diversity ↑ → Segregation ↑ (positive correlation)
- Immigrants and Seniors: Sorting dominates—more diversity, more segregation
- Women: Stable — or (s)low integration despite stable representation?
Trajectories
| Group | Segregation 1996 | Segregation 2022 | Change |
|---|---|---|---|
| Women | 0.29 | 0.26 | −10.3% |
| Seniors | 0.03 | 0.07 | +133.3% |
| Immigrants | 0.06 | 0.17 | +183.3% |
Despite decades of gender equality policy, female segregation barely moved. For immigrants and seniors, segregation increased as their labor force share grew.
Interpretation
Results align with segmented labor market theory: albeit not to extreme extents, demographic groups increasingly concentrate in distinct organizational niches
Increased national diversity does not translate to workplace integration
Instead, growth in group size appears to deepen segregation as groups are channeled into separate segments of the labor market
Bottom line: Pool effects are overwhelmed by sorting effects for immigrants and seniors. The labor market absorbs demographic change by segregating rather than integrating.
(Segregation by sector)
Public sector
| Group | 1996 | 2022 | Change | N 1996 | % 1996 | N 2022 | % 2022 |
|---|---|---|---|---|---|---|---|
| Women | 0.28 | 0.21 | −25.0% | ≈350,000 | 61% | ≈440,000 | 67% |
| Seniors | 0.03 | 0.06 | +100.0% | ≈75,000 | 13% | ≈175,000 | 26% |
| Immigrants | 0.03 | 0.12 | +300.0% | ≈20,000 | 3% | ≈80,000 | 12% |
Private sector
| Group | 1996 | 2022 | Change | N 1996 | % 1996 | N 2022 | % 2022 |
|---|---|---|---|---|---|---|---|
| Women | 0.18 | 0.20 | +11.1% | ≈190,000 | 34% | ≈320,000 | 37% |
| Seniors | 0.03 | 0.17 | +466.7% | ≈50,000 | 9% | ≈190,000 | 22% |
| Immigrants | 0.08 | 0.31 | +287.5% | ≈22,000 | 4% | ≈150,000 | 18% |
H2: Skill-Mediated vs. Residual Sorting
How much segregation is explained by observable skills?
Gender segregation: Horizontal sorting
| Component | Explained |
|---|---|
| Education | 3% |
| Occupation | 23% (28% in 1996) |
Interpretation: Gender segregation operates horizontally—men and women have similar education levels but sort into different occupations and workplaces.
The large residual suggests self-selection into family-friendly workplaces, gendering of work, or gendered recruitment networks.
Immigrant segregation: Vertical sorting
| Component | Explained |
|---|---|
| Education | 56% (31% in 1996) |
| Occupation | 8% |
Interpretation: Immigrant segregation operates vertically—driven by credential gaps and foreign qualification barriers.
The increasing role of education over time suggests generational convergence may reduce segregation. Remaining residual may reflect discrimination or ethnic recruitment networks.
Senior segregation: Mechanisms unrelated to skills?
| Component | Explained |
|---|---|
| Education | 12% (<1% in 1996) |
| Occupation | 8% |
Interpretation: Age-based sorting is largely unrelated to measurable human capital.
Primary mechanisms likely include cohort-specific labor market attachment, technological displacement, or employer stereotypes about older workers.
Group-specific sorting pathways
| Group | Primary pathway | Possible mechanism | Residual |
|---|---|---|---|
| Women | Horizontal | Occupational differentiation | 73% |
| Immigrants | Vertical | Credential gaps | 43% |
| Seniors | Non-skill | Cohort/life decisions | 90% |
H2 supported: Sorting mechanisms differ systematically across groups—occupational for gender, educational for immigrants, and largely non-observable for seniors.
H3: Workplace size
Does segregation decrease with organizational scale?
- All groups: Larger workplaces → lower segregation
- Exception: Gender segregation increases slightly between 500–999 and 1000+ employees
H3 supported: Organizational scale moderates segregation—possibly a reflection of formalized hiring, regulatory pressure, and reduced reliance on network recruitment.
H4: Workplace skill level
Does segregation vary across the skill spectrum?
NOTE: this is preliminar results and robustness test are being done on the workplace skill index measure
H4 preliminary suggestion: Segregation patterns vary systematically by workplace skill level, with distinct shapes for each demographic group reflecting different sorting mechanisms.
Main findings
| Group | Level (2022) | Trend | Implication |
|---|---|---|---|
| Women | Highest (26%) | Stable/declining | Entrenched but not worsening |
| Immigrants | Moderate (17%) | Rising fast (+183%) | Active sorting intensifying |
| Seniors | Lowest (7%) | Rising (+133%) | Emerging problem? |
Main findings
The labor market appear to absorb demographic change by channeling groups into specific workplaces, not by integrating them across workplaces.
Sorting effects dominate pool effects.
Despite Denmark’s tight labor markets, anti-discrimination laws, and dramatic demographic change—conditions maximally favorable to integration—workplace segregation has increased for immigrants and seniors and remains relatively stable for women.
Denmark’s experience suggests a sobering conclusion:
Demographic diversification of the labor force does not automatically produce workplace integration. Without targeted intervention addressing group-specific sorting mechanisms, increased diversity may paradoxically deepen segregation.
Theoretical contributions
- Pool effects have limits: Even under ideal conditions, competitive markets do not eliminate segregation
- Sorting is heterogeneous: Different demographic groups experience fundamentally different mechanisms
- Context matters: Workplace characteristics moderate segregation dynamics
Future Directions
- Within-workplace segregation: Do large workplaces integrate or segregate internally? (see Diop and Larsen, 2024)
- Intersectionality: How do multiple group memberships interact?
- Mechanisms: Can we directly identify discrimination vs. preferences/self-selection?
Current struggles …
Index construction
Interpretation of “strength”
What should be the baseline or point of comparison?